A DEVICETERRA PRODUCT

Technology for emerging-market businesses and institutions.

FREE · PRIVATE · EVIDENCE-FIRST

Can this computer
run local AI?

Describe the device. MamiLens will estimate a safe memory budget, shortlist models and recommend a starting engine. It will not invent a speed result.

Private by designYour hardware choices stay in this browser unless you choose to share the result.

1 · DESCRIBE THE DEVICE

Use the real specifications.

If you do not know the GPU, choose “not sure.” MamiLens will make the safer CPU-only assumption.

Operating systemInstalled system / unified RAMCPU classGraphics hardwareExact GPU modelVRAM per GPUNumber of GPUsMain jobFree space on installation driveLeave blank if unknown. Storage will remain unchecked.Working context

Advanced calculation optionsQuantization, cache and simultaneous users

Model quantizationQ4 is the simple default. Larger quantizations need more memory.KV cache precisionCheck that the chosen engine supports cache quantization before changing this.People active at onceThis estimates simultaneous requests, not the number of accounts.

2 · CONSERVATIVE RESULT

37 curated profiles fit safely

These models pass the current memory and input rules. Start with the first recommendation and benchmark it.

Installed RAM32 GB

−

System reserve6 GB

Model working budget26 GB

RECOMMENDED ENGINE START

LM Studio or Ollama

A CPU-first setup is the safest assumption when dedicated GPU support is unknown.

More CPU cores can help, but memory bandwidth and engine settings still matter.

Recommended to test

Qwen 2.5 14B

Qwen 2.5 14B is a curated dense profile for general, writing, coding workloads.

Estimated model file: 9 GB
Estimated total RAM: 17 GB
Headroom: 15 GB
Acceleration: CPU / RAM
Speed: Benchmark required
Installation space estimate: 14 GB · not checked
Evidence: Calculated estimate

Open model profile →ollama run qwen2.5:14b

Recommended to test

Phi-4 14B

A compact Microsoft model aimed at strong reasoning, mathematics and instruction following.

Estimated model file: 9.1 GB
Estimated total RAM: 17 GB
Headroom: 15 GB
Acceleration: CPU / RAM
Speed: Benchmark required
Installation space estimate: 14.1 GB · not checked
Evidence: Calculated estimate

Open model profile →ollama run phi4:14b

Recommended to test

Qwen 3 14B

Qwen 3 14B is a curated dense profile for general, coding, reasoning workloads.

Estimated model file: 9.3 GB
Estimated total RAM: 18 GB
Headroom: 14 GB
Acceleration: CPU / RAM
Speed: Benchmark required
Installation space estimate: 14.3 GB · not checked
Evidence: Calculated estimate

Open model profile →ollama run qwen3:14b

Recommended to test

Mistral Nemo 12B

Mistral Nemo 12B is a curated dense profile for general, writing, tools workloads.

Estimated model file: 7.1 GB
Estimated total RAM: 15 GB
Headroom: 17 GB
Acceleration: CPU / RAM
Speed: Benchmark required
Installation space estimate: 12.1 GB · not checked
Evidence: Calculated estimate

Open model profile →ollama run mistral-nemo:12b

Recommended to test

Gemma 4 12B QAT

Gemma 4 12B QAT is a curated dense profile for general, writing, reasoning workloads.

Estimated model file: 7.2 GB
Estimated total RAM: 17 GB
Headroom: 15 GB
Acceleration: CPU / RAM
Speed: Benchmark required
Installation space estimate: 12.2 GB · not checked
Evidence: Calculated estimate

Open model profile →ollama run gemma4:12b-it-qat

Recommended to test

Gemma 3 12B

Gemma 3 12B is a curated dense profile for general, writing, reasoning workloads.

Estimated model file: 8.1 GB
Estimated total RAM: 18 GB
Headroom: 14 GB
Acceleration: CPU / RAM
Speed: Benchmark required
Installation space estimate: 13.1 GB · not checked
Evidence: Calculated estimate

Open model profile →ollama run gemma3:12b

Important:

The catalogue covers 151 current, workstation, mobile, server and important legacy GPU profiles. An unknown GPU or multi-GPU system is never treated as verified acceleration. Every speed range is a bandwidth-based calculation, not a measured result. Exact model files, context, runtime version, background applications and cooling can still change the result.

See every calculation and evidence label →Report what happened on your computer →

POPULAR CHECKED DEVICES

Start with a common hardware profile.

These permanent pages use the same conservative rules as the calculator. Choose the closest setup, then enter your exact device above.

8GB RAM, no GPUCan 8GB RAM run local AI?16GB RAM, no GPUWhat local AI can run on 16GB RAM?32GB RAM, no GPUWhich local AI models fit 32GB RAM?16GB RAM, 8GB VRAMWhat local AI runs with 8GB VRAM?32GB RAM, 12GB VRAMWhat local AI runs with 12GB VRAM?Apple Silicon, 16GBCan a 16GB Mac run local AI?Apple Silicon, 32GBWhich local AI models fit a 32GB Mac?8GB RAM for codingWhich coding model can run on 8GB RAM?16GB RAM for imagesWhich local vision model fits 16GB RAM?RTX 3060 12GBWhat LLM can an RTX 3060 12GB run?RTX 4060 8GBWhat LLM can an RTX 4060 run?RTX 4090 24GBWhat LLM can an RTX 4090 run?RX 7900 XTX 24GBCan an RX 7900 XTX run local AI?Intel Arc A770 16GBCan Intel Arc A770 run a local LLM?Apple M1, 16GBWhat local AI can an M1 Mac run?M2 Pro, 32GBWhat local AI can an M2 Pro Mac run?M3 Max, 64GBWhat local AI can an M3 Max Mac run?M4 Pro, 48GBWhat local AI can an M4 Pro Mac run?32GB RAM for codingWhich coding LLM runs on 32GB RAM?32GB private documentsWhat local AI can search private documents on 32GB RAM?64GB Linux team serverWhat local LLM can serve a small team on 64GB RAM?

WHAT THE RESULT MEANS

A safe shortlist, not a speed guarantee.

Architecture-aware memory

The calculator reserves model weights, runtime memory, operating system memory, model-specific KV cache and extra multimodal memory.

Verified GPU path

Reported VRAM is counted only when the exact GPU family and backend support are known. Unknown and multi-GPU systems require validation.

Practical CPU limit

A model that loads can still be too slow. Large CPU-only plans are marked as likely slow instead of being called a strong match.